Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Labeled Datasets”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 253 records · Page 14

Automated Labeling of Electron Microscopy Images Using Deep Learning

© 2018 IEEE. Searching for scientific data requires metadata providing a relevant context. Today, generating metadata is a time and labor intensive manual process that is often neglected, and important datasets are not accessible through search. We investigate the use of machine learning to generalize metadata from a subset of labeled data, thus increasing the availability of meaningful metadata for search. Specifically, we consider electron microscopy images collected at the National Center for Electron Microscopy (NCEM) at the Lawrence Berkeley National Laboratory (LBNL) and use deep learning to discern characteristics from a small subset of labeled images and transfer labels to the entire image corpus. Relatively small training set sizes and a minimum resolution of 512\times 512 pixels required by the application domain pose unique challenges. We overcome these challenges by using a simple yet powerful convolutional network architecture that limits the number of free parameters to lower the required amount of computational power and reduce the risk of overfitting. We achieve a classification accuracy of approximately 80% in discerning between images recorded in two operating modes of the electron microscope-transmission electron microscopy (TEM) and scanning transmission electron microscopy (STEM). We use transfer learning-i.e., re-using the pre-trained convolution layers from the TEM vs. STEM classification problem-to generalize labels and achieve an accuracy of approximately 70 % despite current experiments being limited to small training data sets. We present these predictions as suggestions to domain scientists through a web-based tool to accelerate the labeling process with the goal of further validating our approach and improving the accuracy of automatically created labels.

Weber, Gunther H↗

Unclassified Data Export from the Containment Database

The attached dataset is an unclassified subset of data copied from the CDB for release to a wider audience. It includes only the clearly unclassified columns from each of the tables in the CDB, and only the rows related to announced tests. Also included is a glossary that lists each of the tables and columns included in the export with its definition, historical labels, unit of measure, and reference information. Our objective is to make this unclassified dataset available for querying on an unclassified server through a repeatable process.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Topological Analysis of The SPOKE Graph

The SPOKE graph [2, 6] is a sparse decorated semantic graph representing a collection of knowledge collected in many scientific databases from the fields of healthcare, biochemistry, chemistry, biology, et cetera. This knowledge graph is stored as a relational dataset decorated with metadata on each constituent vertex and edge. Formally, the graph is G(V, E, D), where V is a set of n vertices V := {1, ..., n} and edges of the form (i, j) ϵ E for i, j ϵ V, and table D that for any item in V υ E stores unstructured data such as vertex/edge type, nature of a relationship, et cetera. D(i) = {data involving vertex i ϵ V}, and D(i, j) = {data involving edge (i, j) ϵ E}. Here, we treat the graph as undirected in the sense that a direct relationship for (i, j) causes a (possibly opposite) reverse direct relationship for (j, i). The SPOKE graph G(V, E, D) is formed by processing a collection of relational datasets from medicine, chemistry, and biology, connecting many entities. Here, we analyze an instance from 2019, Spoke-20190707, where a graph file contains 6.16M edges and associated metadata and a vertex file contains 2.15M vertices and the associated metadata. There are 12 different types of vertex entities; all edge types used are implicit (see §2). There is other metadata in D on edges and vertices, but we just use the topology and the vertex labels in this report. SPOKE is growing as more knowledge is gained and more datasets are added. SPOKE is likely to grow 10x during the next phase of this project, and we therefore would like to consider topoligical analysis techniques that are scalable to several orders of magnitude larger than the current dataset (say >1B edges).

59 BASIC BIOLOGICAL SCIENCES↗

TPCpp-10M: Simulated proton-proton collisions in a time projection chamber for AI foundation models

Scientific foundation models hold great promise for advancing nuclear and particle physics by improving analysis precision and accelerating discovery. Yet, progress in this field is often limited by the lack of openly available large scale datasets, as well as standardized evaluation tasks and metrics. Furthermore, the specialized knowledge and software typically required to process particle physics data pose significant barriers to interdisciplinary collaboration with the broader machine learning community. This work introduces a large, openly accessible dataset of 10 million simulated proton-proton collisions, designed to support self-supervised training of foundation models. To facilitate ease of use, the dataset is provided in a common NumPy format. In addition, it includes 70,000 labeled examples spanning three well defined downstream tasks: track finding, particle identification, and noise tagging, to enable systematic evaluation of the foundation model's adaptability. The simulated data are generated using the Pythia Monte Carlo event generator at a center of mass energy of $\sqrt{s}$ = 200 GeV and processed with Geant4 to include realistic detector conditions and signal emulation in the sPHENIX Time Projection Chamber at the Relativistic Heavy Ion Collider, located at Brookhaven National Laboratory. This dataset resource establishes a common ground for interdisciplinary research, enabling machine learning scientists and physicists alike to explore scaling behaviors, assess transferability, and accelerate progress toward foundation models in nuclear and high energy physics. The complete simulation and reconstruction chain is reproducible with the sPHENIX software stack. All data and code locations are provided under Data Accessibility.

Data Analysis, Statistics and Probability (physics↗

A hybrid machine-learning approach for analysis of methane hydrate formation dynamics in porous media with synchrotron CT imaging

Fast multi-phase processes in methane hydrate bearing samples pose a challenge for quantitative micro-computed tomography study and experiment steering due to complex tomographic data analysis involving time-consuming segmentation procedures. This is because of the sample's multi-scale structure, which changes over time, low contrast between solid and fluid materials, and the large amount of data acquired during dynamic processes. Here, a hybrid approach is proposed for the automatic segmentation of tomographic data from time-resolved imaging of methane gas-hydrate formation in sandy granular media, which includes a deep-learning 3D U-Net model. To prepare a training dataset for the 3D U-Net, a technique to automate data labeling based on sample-specific information about the mineral matrix immobility and occasional fluid movement in pores is proposed. Automatic segmentation allowed for studying properties of the hydrate growth in pores, as well as dynamic processes such as incremental flow and redistribution of pore brine. Results of the quantitative analysis showed that for typical gas-hydrate stability parameters (100 bar methane pressure, 7°C temperature) the rate of formation is slow (less than 1% per hour), after which the surface area of contact between brine and gas increases, resulting in faster formation (2.5% per hour). Hydrate growth reaches the saturation point after 11 h of the experiment. Finally, the efficacy of the proposed segmentation scheme in on-the-fly automatic data analysis and experiment steering with zooming to regions of interest is demonstrated.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Enhancing atlas based segmentation with multiclass linear classifiers

Purpose: To present a method to enrich atlases for atlas based segmentation. Such enriched atlases can then be used as a single atlas or within a multiatlas framework. Methods: In this paper, machine learning techniques have been used to enhance the atlas based segmentation approach. The enhanced atlas defined in this work is a pair composed of a gray level image alongside an image of multiclass classifiers with one classifier per voxel. Each classifier embeds local information from the whole training dataset that allows for the correction of some systematic errors in the segmentation and accounts for the possible local registration errors. The authors also propose to use these images of classifiers within a multiatlas framework: results produced by a set of such local classifier atlases can be combined using a label fusion method. Results: Experiments have been made on the in vivo images of the IBSR dataset and a comparison has been made with several state-of-the-art methods such as FreeSurfer and the multiatlas nonlocal patch based method of Coupé or Rousseau. These experiments show that their method is competitive with state-of-the-art methods while having a low computational cost. Further enhancement has also been obtained with a multiatlas version of their method. It is also shown that, in this case, nonlocal fusion is unnecessary. The multiatlas fusion can therefore be done efficiently. Conclusions: The single atlas version has similar quality as state-of-the-arts multiatlas methods but with the computational cost of a naive single atlas segmentation. The multiatlas version offers a improvement in quality and can be done efficiently without a nonlocal strategy.

60 APPLIED LIFE SCIENCES↗

GeoAI advances in specific landform mapping

Landform mapping (also referred to as geomorphology or geomorphometry) can be divided into two domains: general and specific (Evans 2012). Whereas general landform mapping categorizes all elements of the study area into landform classes, such as ridges, valleys, peaks, and depressions, the mapping of specific landforms requires the delineation (even if fuzzy) of individual landforms. The former is mainly driven by physical properties such as elevation, slope, and curvature. The latter, however, must consider the cognitive (human) reasoning that discriminates individual landforms in addition to these physical properties (Arundel and Sinha 2018). Both mapping forms are important. General geomorphometry is needed to understand geological and ecological processes and as boundary layer input to climate and environmental models. Specific geomorphometry supports such activities as disaster management and recovery, emergency response, transportation, and navigation. In the United States, individual landforms of interest are named in the U.S. Geological Survey (USGS) Geographic Names Information System, a point dataset captured specifically to digitize geographic names from the USGS Historical Topographic Map Collection (HTMC). Named landform extent is represented only by the name placement in the HTMC. Recent work has investigated CNN-based deep learning methods to capture these extents in machine-readable form. These studies first relied on physical properties (Arundel et al. 2020) and then included the HTMC as a band in RGB images in limited testing (Arundel et al. 2023). Results from the HTMC dataset surpassed those using just physical properties and using the HTMC alone performed best due to the hillshading and elevation (contour) data incorporated into the topographic maps. However, results fell short of an operational capacity to map all named landforms in the United States. Thus, our current work expands upon past research by focusing on the HTMC and physical information as inputs and the named landform label extents. Specifically, we propose to leverage pre-trained foundation models for segmentation and optical character recognition (OCR) models to jointly map landforms in the United States. Our approach aims to bridge the disparities among the independent information sources to facilitate informed decision-making. The modeling pipeline performs (1) segmentation using the physical information and (2) information extraction using OCR, in parallel. Then a computer vision approach merges the two branches into a labeled segmentation. References: Arundel, Samantha T., Wenwen Li, and Sizhe Wang. 2020. “GeoNat v1.0: A Dataset for Natural Feature Mapping with Artificial Intelligence and Supervised Learning.” Transactions in GIS 24 (3): 556–72. https://doi.org/10.1111/tgis.12633. Arundel, Samantha T, and Gaurav Sinha. 2018. “Validating GEOBIA Based Terrain Segmentation and Classification for Automated Delineation of Cognitively Salient Landforms BT - Proceedings of Workshops and Posters at the 13th International Conference on Spatial Information Theory (COSIT 2017).” In Proceedings of Workshops and Posters at the 13th International Conference on Spatial Information Theory (COSIT 2017), Lecture Notes in Geoinformation and Cartography, edited by Paolo Fogliaroni, Andrea Ballatore, and Eliseo Clementini, 9–14. Cham: Springer International Publishing. Arundel, Samantha T., Gaurav Sinha, Wenwen Li, David P. Martin, Kevin G. McKeehan, and Philip T. Thiem. 2023. “Historical Maps Inform Landform Cognition in Machine Learning.” Abstracts of the ICA 6 (August): 1–2. https://doi.org/10.5194/ica-abs-6-10-2023. Evans, Ian S. 2012. “Geomorphometry and Landform Mapping: What Is a Landform?” Geomorphology 137 (1): 94–106. https://doi.org/10.1016/j.geomorph.2010.09.029.

machine learning↗

Few-shot Learning for Post-disaster Structure Damage Assessment

Automating post-disaster damage assessment with remote sensing data is critical for faster surveys of structures impacted by natural disasters. One significant obstacle to training state-of-the-art deep neural networks to support this automation is that large quantities of labelled data are often required. However, obtaining those labels is particularly unrealistic to support post-disaster damage assessment in a timely manner. Few-shot learning methods could help to mitigate this by reducing the amount of labelled data required to successfully train a model while achieving satisfactory results. To this end, we explore a feature reweighting method to the YOLOv3 object detection architecture to achieve few-shot learning of damage assessment models on the xBD dataset. Our results show that the feature reweighting approach yield improved mAP over the baseline with significantly fewer labelled samples. In addition, we use t-SNE to analyze the class-specific reweighting vectors generated by the reweighting module in order to evaluate their inter-class and intra-class similarity. We find that the vectors form clusters based on class, and that these clusters overlap with visually similar classes. Those results show the potential to employ this few-shot learning strategy for rapid damage assessment with post-event remote sensing images.

Bowman, Jordan↗

BCARS Simulated Phantom Dataset for Evaluation of Processing Pipelines

Broadband coherent anti-Stokes Raman scattering (BCARS) microscopy is a powerful label-free biological imaging technique, but the raw signal requires careful processing. The vibrationally resonant (Raman) fingerprint signal is usually small compared with instrumental noise sources and the nonresonant background (NRB) inherent in the BCARS signal. Fortunately, the NRB exhibits a systematic phase relationship with the coherent Raman response, acting as a heterodyne amplifier for the weak fingerprint signal. Due to this heterodyne effect, the Raman response can be recovered quantitatively and invariantly across different instruments, provided the NRB shape is known. Even with heterodyne amplification, the amplitudes of fingerprint signal components are often comparable to system noise. Singular value decomposition (SVD), which utilizes spatial information, is often employed for additional noise filtering. Consequently, finding optimal processing parameters to properly distinguish the NRB and Raman responses and suppress noise in the complex BCARS signal requires a reference system that realistically represents the spectral and spatial properties of BCARS signals obtained from biological samples. We present a digital tissue phantom that meets these criteria as a tool for testing candidate signal processing pipelines. The digital phantom is generated with simulated hyperspectral Raman images having system-specific noise and background characteristics. Here, we analyze phantom datasets with differing background and signal-to-noise conditions to evaluate their impact on the performance of multiple signal processing pipelines. Specifically, we investigate the application of a Butterworth filter-based routine to directly estimate the NRB from the BCARS signal. Additionally, we evaluate a Lorentzian wavelet transform as an alternative to the Hilbert transform for extracting the Raman spectrum from the BCARS signal. While we demonstrate this phantom for BCARS, it can be used for any spectroscopic Raman imaging approach.

Dixon, Jessica Z. [Georgia Institute of Technology↗

Development of Machine Learning Algorithms to Segment and Study Images of Astromaterial Samples

Introduction: Micrometer-scale chemical analyses of chondritic meteorites and mission-returned asteroid samples can reveal details of the physical and chemical processes operating in the early solar system, including processes that gave rise to planets, moons, and minor bodies. These primitive astromaterials are comprised of chondrules, calcium- and aluminum-rich inclusions (CAI), and many other silicates, oxides, metals, sulfides, and fine-grained materials. The chemical and mineralogical complexity of these samples, vast populations of different components, and heterogeneity across mm to km scales, all limit our understanding of the origin and evolution of these materials. Here, we describe recent efforts to use machine learning techniques to automate the segmentation of chemical maps of chondritic meteorites, designed to aid studies of asteroid samples returned by spacecraft. By automating the task of segmentation it will become possible to rapidly analyze and interpret the sizes, shapes, mineralogy, chemistry, and other properties of every chondrule, calcium- and aluminum-rich inclusion (CAI) and other clast within and between asteroid samples. Sample return missions significantly accelerate and heighten the need to develop such new data analysis techniques, and associated data repositories. Techniques: Neural networks require abundant training data, i.e. images which have been segmented by a human user. We have manually segmented data available from previous petrologic and chemical work at NASA Johnson Space Center and the American Museum of Natural History [1-4]. These data were derived from energy- and wavelength-dispersive X-ray spectroscopy (EDS, WDS) mapping of samples from many chondrite groups. The Deeplabv3+ [5] neural network architecture was trained on human-labeled masks and used to create machine-labeled masks. Several different algorithms were investigated, with inputs ranging from common RGB image formats through to hyperspectral datasets, with raw data comprising greyscale maps of Mg, Ca, and Al, with or without Si, Fe, Ti for both EDS and WDS data, and extending to other elements in EDS only. Each greyscale image was paired with a binary mask for each labelled particle type. Results: The trained algorithms can segment (Fig 1), classify, and measure the dimensions of thousands of particles in chemical maps of a standard 1-inch round petrographic section in seconds to minutes, rather than many hours needed by a human. Accuracy of the algorithms varied from chondrite to chondrite and across particle types. Further results and details of the algorithms will be presented at the workshop. Future directions: Machine learning has the potential to revolutionize our understanding of complex particle populations contained within primitive astromaterial, with segmentation being a critical first step. Example applications include better understanding of particle transport, nebular reservoirs, parent body accretion, and a deeper understanding of the relationships between particle populations and bulk rock elemental and isotopic compositions. In addition to benefits that machine learning can bring to individual researchers, building a community data repository of thousands to millions of particles across hundreds of samples will open up many other possibilities. For example, with a large enough dataset it will be possible to search for exceptionally closely matching particles across disparate samples. Such a capability would enable a single CAI from OSIRISREx or Hayabusa/II samples to be matched to chondritic CAIs that exhibit near-identical size, texture, and mineralogy, down to the level of similar core phenocrysts, zonation, and rim sequences. Such comparative analyses will help to disentangle precursor chemistry, chronology, gas/dust reservoirs during heating, and accretion. Such an endeavor would be impossible without machine learning and a large community data repository of astromaterial chemical/mineralogic maps.

Machine Learning↗

AI4MARS: A Dataset for Terrain-Aware Autonomy on Mars

Deep learning has quickly become a necessity for selfdriving vehicles on Earth. In contrast, the self-driving vehicles on Mars, including NASA’s latest rover, Perseverance, which is planned to land on Mars in February 2021, are still driven by classical machine vision systems. Deep learning capabilities, such as semantic segmentation and object recognition, would substantially benefit the safety and productivity of ongoing and future missions to the red planet. To this end, we created the first large-scale dataset, AI4Mars, for training and validating terrain classification models for Mars, consisting of ~326K semantic segmentation full image labels on 35K images from Curiosity, Opportunity, and Spirit rovers, collected through crowdsourcing. Each image was labeled by ~10 people to ensure greater quality and agreement of the crowdsourced labels. It also includes ~1.5K validation labels annotated by the rover planners and scientists from NASA’s MSL (Mars Science Laboratory) mission, which operates the Curiosity rover, and MER (Mars Exploration Rovers) mission, which operated the Spirit and Opportunity rovers. We trained a DeepLabv3 model on the AI4Mars training dataset and achieved over 96% overall classification accuracy on the test set. The dataset is made publicly available.1

Ono, Hiro↗

Machine learning for reactor power monitoring with limited labeled data

Real-time reactor power monitoring is critical for a variety of nuclear applications, spanning safety, security, operations, and maintenance. While machine learning methods have shown promise in monitoring reactor power levels, there is limited research on their efficacy in label-starved environments. The goal of this work is to assess the feasibility of classifying nuclear reactor power level using multisource data in scenarios with limited labels. Data were collected using low-resolution multisensors at four nuclear reactor facilities: two large research reactors and two TRIGA reactors. Within each pair, one reactor dataset served as the source and the other as the target in a transfer learning paradigm. Twenty-three supervised models were trained on labeled sequences of magnetic field and acceleration data from each of the target sites. Self-learning and transfer learning methods were applied to the top performing models to assess their classification performance with increasing amounts of labeled data. While reactor power level classification was achieved with a Matthews Correlation Coefficient of up to 0.739 ± 0.003 and 0.622 ± 0.009 with only 400 sequences per power state for the large research reactor and TRIGA target sites, respectively, self-learning and transfer learning leveraging source site data did not improve target classification performance. These findings suggest that alternative methods, such as higher sensitivity sensors, digital twins, or the use of physics-informed models, are required to enable high-performance classification in machine learning approaches to reactor monitoring with a dearth of target ground truth.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

MERIT-Plus Dataset: Delineation of endorheic basins in 5 and 15 min upscaled river networks

Endorheic drainage basins, those inland basins not connected directly to ocean, are essential for hydrological modeling of global and regional water balances, land surface water storage, gravity anomalies, sea level rise, etc. Within many hydrological model frameworks, river basins are defined by digital river networks through their flow direction and connectivity datasets. Here we present an improvement to gridded flow direction data and its derivatives produced from upscaled global 5 and 15 arc minute MERIT networks. We explicitly label endorheic and exorheic drainage basins and alter the delineation of endorheic basins by merging small inland watersheds to the adjacent host basins. The resulting datasets have a significantly reduced number of endorheic basins while preserving the total land portion of those basins since most of the merged catchments were inside other larger endorheic areas. We developed and present here the endorheic basin delineation method. This method performs an analysis of the contributing river and basin geometry relative to the location of the flow end point (i.e. potential endorheic lake), proximity of the latter to the drainage basin boundary and the elevation difference between the basin's lowest point and potential spillover location at the basin boundary. The new digital river network was validated using the University of New Hampshire Water Balance Model by comparing the water balance of endorheic inland depressions with modeled accumulation of water in their inland lakes based on the observed historical climate drivers used by WBM.

Earth Systems↗

MERIT-Plus Dataset: Delineation of endorheic basins in 5 and 15 min upscaled river networks

Endorheic drainage basins, those inland basins not connected directly to ocean, are essential for hydrological modeling of global and regional water balances, land surface water storage, gravity anomalies, sea level rise, etc. Within many hydrological model frameworks, river basins are defined by digital river networks through their flow direction and connectivity datasets. Here we present an improvement to gridded flow direction data and its derivatives produced from upscaled global 5 and 15 arc minute MERIT networks. We explicitly label endorheic and exorheic drainage basins and alter the delineation of endorheic basins by merging small inland watersheds to the adjacent host basins. The resulting datasets have a significantly reduced number of endorheic basins while preserving the total land portion of those basins since most of the merged catchments were inside other larger endorheic areas. We developed and present here the endorheic basin delineation method. This method performs an analysis of the contributing river and basin geometry relative to the location of the flow end point (i.e. potential endorheic lake), proximity of the latter to the drainage basin boundary and the elevation difference between the basin's lowest point and potential spillover location at the basin boundary. The new digital river network was validated using the University of New Hampshire Water Balance Model by comparing the water balance of endorheic inland depressions with modeled accumulation of water in their inland lakes based on the observed historical climate drivers used by WBM.

Earth Systems↗

MERIT-Plus Dataset: Delineation of endorheic basins in 5 and 15 min upscaled river networks

Endorheic drainage basins, those inland basins not connected directly to ocean, are essential for hydrological modeling of global and regional water balances, land surface water storage, gravity anomalies, sea level rise, etc. Within many hydrological model frameworks, river basins are defined by digital river networks through their flow direction and connectivity datasets. Here we present an improvement to gridded flow direction data and its derivatives produced from upscaled global 5 and 15 arc minute MERIT networks. We explicitly label endorheic and exorheic drainage basins and alter the delineation of endorheic basins by merging small inland watersheds to the adjacent host basins. The resulting datasets have a significantly reduced number of endorheic basins while preserving the total land portion of those basins since most of the merged catchments were inside other larger endorheic areas. We developed and present here the endorheic basin delineation method. This method performs an analysis of the contributing river and basin geometry relative to the location of the flow end point (i.e. potential endorheic lake), proximity of the latter to the drainage basin boundary and the elevation difference between the basin's lowest point and potential spillover location at the basin boundary. The new digital river network was validated using the University of New Hampshire Water Balance Model by comparing the water balance of endorheic inland depressions with modeled accumulation of water in their inland lakes based on the observed historical climate drivers used by WBM.

Earth Systems↗

MERIT-Plus Dataset: Delineation of endorheic basins in 5 and 15 min upscaled river networks

Endorheic drainage basins, those inland basins not connected directly to ocean, are essential for hydrological modeling of global and regional water balances, land surface water storage, gravity anomalies, sea level rise, etc. Within many hydrological model frameworks, river basins are defined by digital river networks through their flow direction and connectivity datasets. Here we present an improvement to gridded flow direction data and its derivatives produced from upscaled global 5 and 15 arc minute MERIT networks. We explicitly label endorheic and exorheic drainage basins and alter the delineation of endorheic basins by merging small inland watersheds to the adjacent host basins. The resulting datasets have a significantly reduced number of endorheic basins while preserving the total land portion of those basins since most of the merged catchments were inside other larger endorheic areas. We developed and present here the endorheic basin delineation method. This method performs an analysis of the contributing river and basin geometry relative to the location of the flow end point (i.e. potential endorheic lake), proximity of the latter to the drainage basin boundary and the elevation difference between the basin's lowest point and potential spillover location at the basin boundary. The new digital river network was validated using the University of New Hampshire Water Balance Model by comparing the water balance of endorheic inland depressions with modeled accumulation of water in their inland lakes based on the observed historical climate drivers used by WBM.

Earth Systems↗

Hyper Parameter Tuning in Neural Optical Image Categorizer for the E-log (NOICE

The Fermilab Accelerator Division Electronic logbook (E-log) is a record of all the activities and events in the Division for the past 10 years and more. The Elog search function is a valuable resource and the institutional memory of the accelerator complex. About 300,000 files are stored in the E-log, of which the vast majority are images attached to entries and comments. The visual information contained in the images is not presently searchable. The goal of team NOICE (Neural Optical Image Categorizer for the E-log) was to design a neural network able to produce label categories for these images for use by searches. The group developed a dataset and trained a convolutional neural network (CNN) with optimized hyperparameter. Final performance metrics are presented.

Njekeu, Terence Franck↗

Towards an Aviation Large Language Model by Fine-tuning and Evaluating Transformers

In the aviation domain, there are many applications for machine learning and artificial intelligence tools that utilize natural language. For example, there is a desire to know the commonalities in written safety reports such as voluntary post incidents reports or create more accurate transcripts of air traffic management conversations. Another use-case is the possibility of extracting airspace procedures and constraints currently written in documents such as Letters of Agreement (LOA) which is used as the evaluation case in this paper. These applications can benefit from the use of state-of-the-art Natural Language Processing (NLP) techniques when adapted to the language/phraseology specific to the aviation domain. This paper evaluates the viability of transferring pre-trained large language models to the aviation domain by adapting transformer based models using aviation datasets. This paper utilized two datasets to adapt a ‘Robustly Optimized Bidirectional Encoder Representations from Transformers Approach’ (RoBERTa) model and two down-stream classification tasks to assess its performance. These datasets are all built upon Letters of Agreement which are Federal Aviation Administration (FAA) documents that formalize airspace operations across the national airspace system. The first two datasets are used for the adaptation of RoBERTa to the aviation domain and were of different sizes to assess the number of documents needed to adapt to the aviation domain. They contain many examples of ‘aviation English’ using domain specific terminology and phrasing which serves as a representative basis to perform the unsupervised adaptation. The second dataset is a separate set of LOA documents with two sets of classification labels to be used for evaluation; one at the document level and one at the line level. These down-stream evaluations allowed the measurement of improvement by adapting RoBERTa. The accuracy increased by 4-6% on both tasks and the F1 score on the class of interest increased by 4-8% from the adaptation.

Air Traffic Management↗